Add expanded anti-rationalization patterns

[mickeylorenzini]

Problem

Harden has anti-patterns that describe audit failures, but not patterns that detect agent self-deception during execution. Agents rationalize skipping steps with plausible-sounding excuses that slip past current checks.

Examples of agent rationalization:

• "I already tested it manually" → skipped automated verification
• "This is straightforward enough to not need tests" → skipped TDD
• "The output looks reasonable" → no oracle verification
• "I'll come back to this later" → deferred and forgotten

Proposed Solution

Add explicit anti-rationalization patterns to harden's audit that catch:

• Claims of completed work without evidence (logs, test output, screenshots)
• Substitution of easier tasks for harder required tasks
• Deferral language that indicates skipped steps
• "Looks good/reasonable" assessments without comparison to an independent source

Likely Affected Files/Modules

• SKILL.md — new patterns in anti-pattern section or as a detection dimension within existing passes

Acceptance Criteria

• At least 5 documented anti-rationalization patterns with examples
• Each pattern has a detection heuristic (what text/behavior triggers it)
• Patterns integrated into appropriate audit pass
• Tested: agent-generated deliverable with known rationalized skip is flagged